This paper proposes a reconstruction method for optimized three-dimensional (3D) morphable model parameters to address the instability of the reconstructed model’s shape expression ability due to the inaccurate detection of facial feature points in the face-image reconstruction method based on the 3D morphable model. First, an improved position map regression network is used to extract and locate the facial feature points accurately and on this basis, the initial model parameters are calculated. The parameters obtained using the regression method are then integrated to obtain the optimised model parameters and improve the model’s accuracy and generalisation capabilities. Finally, to obtain the final face model, the 3D morphable model is optimized. This approach can produce accurate 3D face reconstruction using real faces as experimental data.